1: \begin{abstract}
2: We carry out simulations of gravitationally unstable disks using smoothed particle hydrodynamics(SPH) and
3: the novel Lagrangian meshless finite mass (MFM) scheme in the GIZMO code
4: \citep{Hopkins2015}. Our aim is
5: to understand the cause of the non-convergence of the cooling
6: boundary for fragmentation reported in
7: the literature.
8: We run SPH simulations with two different artificial viscosity implementations, and compare them with MFM,
9: which does not employ any artificial viscosity.
10: With MFM we demonstrate convergence of the critical cooling time scale for fragmentation at $\beta_{crit} \approx 3.$.
11: Non-convergence persists in SPH codes, although it is
12: significantly mitigated with schemes having reduced
13: artificial viscosity such as inviscid SPH (ISPH) \citep{Cullen2010}.
14: We show how the non-convergence problem is caused by
15: artificial fragmentation triggered by excessive
16: dissipation of angular momentum in domains with large velocity derivatives. With increased
17: resolution such domains become more prominent. Vorticity lags behind
18: density due to numerical viscous dissipation in these regions, promoting
19: collapse with longer cooling times. Such effect is shown to be dominant over the competing tendency of artificial viscosity to diminish with increasing resolution.
20: When the initial conditions are first relaxed for several orbits, the flow is more
21: regular, with lower shear and vorticity in non-axisymmetric regions, aiding convergence.
22: Yet MFM is the only method that converges exactly.
23: Our findings are of general interest as numerical dissipation via artificial viscosity or advection errors can also occur in grid-based codes. Indeed for
24: the FARGO code values of $\beta_{crit}$ significantly
25: higher than our converged estimate
26: have been reported in the literature.
27: Finally, we discuss implications for giant planet
28: formation via disk instability.
29:
30:
31:
32:
33:
34: \end{abstract}
35: